Skip to content

Add MGSM-rev2 (MGSM revised version) - #10

Open
Lduignan1 wants to merge 2 commits into
mainfrom
mgsm-rev2
Open

Add MGSM-rev2 (MGSM revised version)#10
Lduignan1 wants to merge 2 commits into
mainfrom
mgsm-rev2

Conversation

@Lduignan1

Copy link
Copy Markdown
Member

Add MGSM-rev2 task (lightonai/mgsm-rev2)

Adds the revised MGSM benchmark for 11 languages (en, es, fr, de, ru, zh, ja, th, sw, bn, te). Test split only and no chain-of-thought field, so the tasks run zero-shot.

Dataset
Paper

Changes

  • default_tasks.py: mgsm-rev2:XX configs so the tasks are runnable flag-free as lighteval|mgsm-rev2:fr|0.
  • multilingual/tasks.py: mgsm_rev2_tasks alongside the existing mgsm_tasks.
  • default_prompts.py: mgsm() now uses line.get("answer") so it tolerates variants that ship no CoT field. Backward compatible with the original MGSM.

@Lduignan1
Lduignan1 requested a review from Jeronymous August 25, 2026 14:31
@Lduignan1 Lduignan1 self-assigned this Aug 25, 2026
@Lduignan1 Lduignan1 added the enhancement New feature or request label Aug 25, 2026

@Jeronymous Jeronymous left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It looks good.
It could be factorized, but the original code was not, so I have no issue with continuing to do copy-pasting. Two small comments below

),
suite=("lighteval",),
hf_repo="lightonai/mgsm-rev2",
hf_revision="1463c6dc7991a8751b8e28e76f6c561b9201eb55",

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why to enfore the revision here?
It's the last one. You want to be defensive in case the dataset is updated in the future?

# reuses the same QA prompt and metric. There is no train split, so it is run zero-shot.
mgsm_rev2_tasks = [
LightevalTaskConfig(
name=f"mgsm_rev2_{language.value}",

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

suggestion: mgsm-rev2_{language.value} ? (dot instead of underscore)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants